Enum Commands

cake-cli/src/main.rs:24–93

#[derive(Subcommand)]
enum Commands {
    /// Run a model locally, as cluster master, or as cluster worker.
    ///
    /// With model: local inference or cluster master.
    /// Without model + --cluster-key: cluster worker (waits for master).
    Run {
        /// Model name or HuggingFace repo (e.g., evilsocket/Qwen3-0.6B).
        /// Omit with --cluster-key to start as a worker node.
        #[arg(id = "model_name")]
        model: Option<String>,
        /// Prompt text (omit for interactive mode)
        #[arg(id = "prompt_text")]
        prompt: Option<String>,
        #[command(flatten)]
        args: Args,
    },
    /// Start an OpenAI-compatible API server.
    Serve {
        /// Model name or HuggingFace repo
        #[arg(id = "model_name")]
        model: String,
        /// API bind address
        #[arg(long = "api", default_value = "0.0.0.0:8080")]
        address: String,
        #[command(flatten)]
        args: Args,
    },
    /// Download a model from HuggingFace Hub
    Pull {
        /// HuggingFace repo ID (e.g., evilsocket/Qwen3-0.6B)
        model: String,
    },
    /// List locally available models
    List,
    /// Interactive chat with a model (local or remote).
    ///
    /// Local mode:  `cake chat <model>` — loads model and runs TUI chat.
    /// Remote mode: `cake chat --server http://host:8080` — connects to API.
    Chat {
        /// Model name or HuggingFace repo for local chat (e.g., Qwen/Qwen3-0.6B).
        /// If omitted, connects to --server instead.
        model: Option<String>,
        /// Server URL for remote chat mode.
        #[arg(long, default_value = "http://localhost:8080")]
        server: String,
        /// Offload expert weights to disk for MoE models larger than available RAM.
        #[arg(long, default_value_t = false)]
        expert_offload: bool,
    },
    /// Delete a cached model.
    Rm {
        /// Model name (e.g., evilsocket/Qwen3-0.6B or just Qwen3-0.6B)
        model: String,
    },
    /// Split a model into per-worker bundles
    Split {
        /// Input model path
        #[arg(long)]